Research On Machine Learning-based Methods For Detecting Surface Defects In The Production Of Metal Materials

China is a big country of iron and steel.At present,China’s iron and steel industry is in a historic critical period from large to strong.The iron and steel industry is carrying out the action plan of intelligent manufacturing and promoting the high-quality development of the industry.Surface defects are an important cause of steel quality.With the rapid development of machine learning theory,the method of inspecting the surface quality of steel and other metal products by human eyes and traditional non-destructive testing technology has gradually been eliminated.More and more steel enterprises are using new technologies such as machine learning and deep learning to achieve production process control and information management,and to identify the surface quality of steel and other metal products.This paper is based on the premise of reviewing a large amount of domestic and international literature.In this paper,on the premise of consulting a large number of domestic and foreign relevant literature,the surface defect detection method in the production of metal materials is studied.Taking hot rolled steel strips as the research object,traditional machine learning and deep learning methods are used to conduct research.1.Machine learning based on manual extraction of features to identify and classify surface defects in strip steel.In order to identify the defects on the surface of hot-rolled strip steel,this paper analyses the target area of defects by considering shape features and texture features,extracts shape features and texture features,fuses the above features and normalises them,and inputs them into the machine learning classifier as feature vectors.In order to improve the classification accuracy,the kernel function and kernel parameters are selected in order to achieve an average classification accuracy of 95.85%.In the comparison experiments,by comparing with the classifier with a single feature,the OAO-SVM classifier proposed in this paper with multiple feature fusion is more accurate and can effectively identify the surface defects of hot-rolled strip steel,with an average correct rate of 8.85% and 13.29%higher than the classifier composed of two single texture features,respectively.2.Research on strip steel surface defect detection method based on YOLOv5 target detection algorithm.Deep learning,an extended field resulting from the continuous development of machine learning,can learn to analyse and extract defect features through convolutional neural networks(CNN),so the target detection algorithm in deep learning can be applied to the detection of defects on strip steel surfaces.As a common detection object in the field of small object detection,surface defects of strip steel usually have problems such as high detection difficulty,low detection accuracy,and small number of defect samples,which results in poor robustness and insufficient generalization ability of the algorithm model.In response to the above problems,this article improves the benchmark model YOLOv5 s through data augmentation,fusion attention mechanism CA,and optimization of output detection head structure(decoupling head).Finally,combined with the above improvement measures,experiments have shown that compared to the benchmark model,the m AP value has increased by 7.9%,and there are fewer false detections and missed detections in the detection result graph.At the same time,the optimized network structure is not too complex and still maintains good real-time performance.The FPS value has only decreased by 7f/s,making it suitable for use in industrial environments of strip steel production.3.In order to better visualise the inspection results and to facilitate the presentation of the actual results of defect classification and inspection,this paper uses the Py Qt5 framework to encapsulate the optimised algorithms based on two aspects of machine learning into a graphical user interface(GUI),which allows the visualisation of the results while saving the inspection data and pictures of the inspection results and giving recommendations on the corresponding troubleshooting measures for different types of defects,in order to facilitate the use of inspection by inspectors in industrial production and to realise process control and information technology.In summary,the two optimization algorithms and visualization design based on machine learning in this paper can be applied to the surface quality detection scene of hot rolled strip,and provide new ideas for surface quality detection of hot rolled strip production enterprises and related metallurgical enterprises.